AI memory stocks have lost more than half their value from a June peak, but the market may still be ahead of the earnings damage.
AI memory stocks fall 51% from June peak

Goldman Sachs’ AI memory basket has fallen 51% from its June high, its steepest drawdown since the ChatGPT-era rally began, as investors unwind one of the most crowded trades in the semiconductor complex. JPMorgan says the selloff is already pricing a memory downcycle that may not hit earnings until the first half of 2027, leaving at least a six-month buffer for suppliers cushioned by long-term contracts and customer prepayments.
That split between price action and fundamentals matters because memory chips sit at the center of the AI buildout. High-bandwidth memory demand helped push the basket more than fourfold higher into June, with Samsung Electronics, SK Hynix, Micron, Kioxia and Western Digital among the names exposed to the surge in data-center spending. Now the market is questioning whether AI capital expenditure has already peaked, a shift that hits not just chipmakers but also the broader valuation framework for the AI trade.
For investors, the selloff is a warning that the market is no longer paying for growth alone. Goldman has said crowded positioning left the sector vulnerable to liquidation, and the latest decline suggests much of that unwinding is underway. A Goldman strategist has also pointed to AI cloud-provider credit spreads at their widest this year, arguing that equity markets may not have fully absorbed financing risk. If credit keeps widening, the pressure could spill from memory stocks into the wider semiconductor rally.
JPMorgan’s more measured stance gives bulls a near-term argument: order visibility is still holding, and earnings cuts may be delayed by prepayments and multi-quarter supply agreements. That makes the next few months more about sentiment and valuation than immediate profit downgrades. But it also means the market is likely to stay highly sensitive to any sign that hyperscaler spending is slowing, contract terms are weakening or AI infrastructure is becoming harder to finance.
Nvidia’s near-term trading and upcoming cloud spending updates will help determine whether the current reset is a healthy correction or the start of a deeper repricing across AI hardware. For now, the message is clear: equities are discounting a downcycle well before the income statement does.
| Entity | Gains | Losses |
|---|---|---|
| Value buyers in semis | ▲Lower entry points | ▼Momentum-driven longs |
| Memory chip suppliers | ▲Near-term earnings buffer | ▼Multiple expansion |
| AI cloud borrowers | ▲Funding still available | ▼Tighter credit conditions |
| Semiconductor indexes | ▲Stabilization if selling eases | ▼Further de-risking if spreads widen |




